Figure 2

Overview of methodology. The framework can be divided into two steps: (1) construction of an integrated functional linkage network (FLN) (2) development of a decision rule for the constructed FLN to optimize functional annotation. Green boxes denote data inputs or products; purple boxes denote actions. In step one six data sources are used as inputs to one or another learning algorithm, to find functionally associated pairs of proteins. The linked proteins identified thus comprise a weighted functional linkage network. In step two, proteins of unknown functions are then annotated based on the collective properties of neighboring annotated proteins, using one or another decision rule. Performance, which we measure by a combination of precision and coverage, is evaluated as described below, using doubly annotated protein pairs.